Nvidia Overtakes Apple as TSMC's Top Customer
Nvidia has surpassed Apple to become TSMC's largest customer, now accounting for 19% of the chipmaker's revenue. The shift highlights the massive demand for AI infrastructure, which is reshaping priorities and capacity allocation within the global semiconductor supply chain.
The battle for cutting-edge silicon is no longer just about nanometers; it's about packaging. Nvidia's surge is powered by massive, multi-die GPUs like the Blackwell B200, which packs 208 billion transistors across two interconnected dies. This complex design is entirely dependent on TSMC's Chip-on-Wafer-on-Substrate (CoWoS) advanced packaging, creating a demand bottleneck that has left capacity sold out well into the future. This shift is reflected directly in TSMC's revenue breakdown. The High-Performance Computing (HPC) segment, which includes Nvidia's AI accelerators, saw a 48% year-over-year revenue increase in 2025, now accounting for 58% of TSMC's total revenue. In contrast, the smartphone segment grew by a much smaller 11%, demonstrating a clear realignment of priorities within the foundry's fabs. In response to this "insane" demand, TSMC is forecasting an 80% compound annual growth rate for its CoWoS capacity and is planning a record capital expenditure of up to $56 billion for 2026. This massive investment is aimed at expanding both advanced packaging and new 2-nanometer production lines, where the next generation of silicon competition will take place. Apple, historically the pioneer of TSMC's latest nodes, is now playing a strategic long game to secure its on-device AI future. The company has reportedly secured more than half of TSMC's initial 2nm production capacity for 2026. This move is crucial for its upcoming A20 and M6 series chips, which will power the next generation of iPhones and Macs. Furthermore, Apple is evolving its hardware integration strategy by moving from its long-standing InFO packaging to a more advanced Wafer-Level Multi-Chip Module (WMCM) for its 2nm chips. This technology will allow for tighter integration of the CPU, GPU, and RAM on a unified wafer platform, a critical step for boosting the performance and efficiency of on-device machine learning applications across its entire product portfolio.
